asr-africa/whisper_DigitalUmuganda_Afrivoice_Shona_86hr_v2

Whisper Small Shona - Beijuka Bruno

This model is a fine-tuned version of openai/whisper-small on the Afrivoice_shona dataset. It achieves the following results on the evaluation set:

• Loss: 0.2732
• Wer: 0.2494
• Cer: 0.0454

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.4587	1.0	3704	0.3029	0.2674	0.0498
0.2209	2.0	7408	0.2808	0.2737	0.0539
0.1485	3.0	11112	0.2871	0.2600	0.0464
0.0954	4.0	14816	0.3082	0.2701	0.0529
0.0567	5.0	18520	0.3315	0.2619	0.0492
0.0328	6.0	22224	0.3654	0.2632	0.0479
0.0205	7.0	25928	0.3952	0.2661	0.0481
0.0146	8.0	29632	0.4262	0.2612	0.0479
0.0116	9.0	33336	0.4367	0.2640	0.0492
0.0096	10.0	37040	0.4722	0.2657	0.0476
0.0081	11.0	40744	0.4748	0.2642	0.0481
0.0075	12.0	44448	0.4790	0.2693	0.0476
0.0069	13.0	48152	0.4983	0.2665	0.0493

Framework versions

• Transformers 4.45.2
• Pytorch 2.2.0+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1